NTC will not show pure linearity! Relationship of resistance with temperature looks like exponential curve: R = A * exp (B / T), with A, B two characteristic parameters and T absolute temperature.

So, you MUST use software linearization. Hardware linearization is also possible, but will not result in +-0.1% accuracy.

NTC is not a good choice, when wanting to achieve real +-0.1% precision. The reason is, that with standard NTC material long term drift is rather poor. After 1000 hours at 25°C drift of R25 can be up to +-1%. After 1000 hours at 125°C drift of R25 can be up to +-3%. After 56 days at 40°C and 90-95% RH drift of R25 can be up to +-3%.

Is 3% much?
Let's have a look at some typical NTC with R25 = 10000 Ohm. At 30°C resistance of NTC is R30 = 8060 Ohm. This represents a relative change of 19.4%. Assuming, that R of NTC changes nearly linearily between 25°C and 30°C, this yields a relative change of 3.88% per °C. So, 3% change of resistance of NTC represents a temperature inaccuracy of about +-0.8°C. That's far away from what you want, namley +-0.1%!

Fortunately, there are high accuracy NTCs available, showing somewhat lower drift. Typical drift after 1000 hours at 25°C is sometimes specified to be about +-0.3%. So, you CAN result after 1000 hours with a drift of less than about +-0.1%. But you have no guarantee! Also, 1000 hours represents a period of only a bit more than one month. But what about drift after half a year? What about drift, when temperature and relative humidity (RH) are rised?

I think, due to rather poor long term drift NTCs will only be good enough for achieving an accuracy of about +-0.5°C...+-1°C. Only if you can get a very very precise type with specified long term drift, better accuracy can be achieved.

In your case I would use PT100.

Kai